Sequence mapping load identification decomposition method based on multi-scale residual neural network

Abstract

The invention relates to the technical field of load identification, in particular to a sequence mapping load identification decomposition method based on a multi-scale residual neural network, which comprises the following steps that: a load identification model is constructed, and the network structure of the load identification model is a multi-scale residual network based on expansion convolution; acquiring an aggregation power data set and an electric appliance level power data set in the public data set, and preprocessing the aggregation power data set and the electric appliance level power data set to obtain a preprocessed training set and a preprocessed test set; and training and testing the load identification model by using the preprocessed training set and test set to obtain a trained load identification model. According to the method, the load identification model of the multi-scale residual network based on the expansion convolution is constructed, so that mixed data features can be learned in a more targeted manner, and the load identification effect of the load identification model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of load identification, and in particular to a sequence mapping load identification and decomposition method based on a multi-scale residual neural network. Background Art

[0002] As household electricity consumption demands become increasingly complex, smart grids and load monitoring technologies have become indispensable. Among them, load monitoring includes intrusive load monitoring (ILM) and non-intrusive load monitoring (NILM). Intrusive load monitoring refers to installing sensors on each of the user's electrical appliances. Although the measurement data obtained through intrusive load monitoring is accurate and reliable, it has defects such as high investment cost, low operability and user acceptance. Non-intrusive load monitoring only sets a measurement point at the entrance of the user's home. With the increasing demand for economy and efficiency of measurement technology, non-intrusive load monitoring, as a cost-effective electricity measurement method, has received more and more attention.

[0003] Existing non-intrusive load monitoring methods mainly include methods based on key signal features and data-driven methods. The former does not require statistical learning of the model and only requires analysis of the signal spectrum characteristics to achieve the best results, while the latter relies on a certain amount of labeled data to learn a representative model, which can be further roughly divided into traditional machine learning, deep learning and online transfer learning.

[0004] For methods based on key signal features, they rely on accurately detecting the switching events of electrical appliances. Signal processing techniques can be used to detect these key features. Among them, electrical features such as current, voltage, and active power are based on the time domain features of the waveform. On the other hand, advanced strong time-varying signal processing algorithms introduce more transform domain features, such as spectrum-based features and wavelet-based features. With sufficient artificial feature description labels, the optimization framework can be applied to non-training methods. However, the generalization ability of the model always depends on prior expert knowledge. This method has obvious defects in the deployment of some large-scale energy management systems.

[0005] For data-driven methods, statistical learning frameworks such as single-label classification and multi-label classification are used. In deep neural networks, better performance can be achieved as the network depth increases, but the network will face degradation problems, including gradient explosion and gradient disappearance. Although some existing models can achieve good load identification performance in simple electrical mixed application scenarios, the identification effect in complex scenarios is not satisfactory. Summary of the invention

[0006] In order to solve the technical problem that the above-mentioned existing non-intrusive load monitoring methods have low identification effect, the purpose of the present invention is to provide a sequence mapping load identification and decomposition method based on a multi-scale residual neural network. The technical solution adopted is as follows: An embodiment of the present invention provides a sequence mapping load identification and decomposition method based on a multi-scale residual neural network, the method comprising the following steps: Constructing a load identification model, wherein the network structure of the load identification model is a multi-scale residual network based on dilated convolution; Obtaining an aggregated power data set and an appliance-level power data set from a public data set, preprocessing the aggregated power data set and the appliance-level power data set to obtain a preprocessed training set and a test set; The load identification model is trained and tested using the preprocessed training set and test set to obtain a trained load identification model.

[0007] Further, the multi-scale residual network based on dilated convolution includes: a dilated convolution residual block; The input of the dilated convolution residual block passes through two convolutional layers and their corresponding normalization layers, dropout layers, and nonlinear activation functions in sequence; then, the original input is added to the temporal output through a shortcut connection to form the final output of the dilated convolution residual block; wherein the two convolutional layers in the dilated convolution residual block share the same dilation rate.

[0008] Furthermore, the calculation formula of the receptive field of the dilated convolution residual block is: ; Where S represents the receptive field of the dilated convolution residual block, N represents the number of dilated convolutions, n represents the number of network layers of dilated convolutions, and k represents the convolution kernel size.

[0009] Furthermore, the obtaining of the aggregated power dataset and the appliance-level power dataset in the public dataset includes: Obtain the UK-DALE public data set, and select the aggregate power data and appliance-level power data of each target appliance in the first preset number of target households according to the same sampling frequency in the UK-DALE public data set; wherein the first preset number of target households are households with the second preset number of appliances of the same type, and the target appliances are appliances of the same type; Determine the time period when the electrical appliances of each target household are turned on. Based on the obtained aggregate power dataset and appliance-level power dataset, use the Electric.get_activation function of non-intrusive load monitoring and the parameters provided by UK-DALE to obtain activation information during the turn-on period, and then obtain a pair of aggregate power sequences and label power sequences corresponding to each time index.

[0010] Further, the preprocessing of the aggregated power data set and the appliance-level power data set to obtain a preprocessed training set and a test set includes: First, the aggregated power data set and the appliance-level power data set are normalized to obtain a normalized aggregated power data set and an appliance-level power data set; Based on the normalized aggregate power data set and appliance-level power data set, a window of preset scale is used to divide the power data that does not contain the activation information of the target appliance from the activation starting point, and the updated aggregate power data set and appliance-level power data set are obtained. The updated appliance-level power data set is used as the preprocessed test set. The updated aggregated power data set is filtered to obtain a filtered aggregated power data set as a preprocessed training set.

[0011] Further, filtering the updated aggregated power data set to obtain the filtered aggregated power data set includes: The updated aggregated power data of the preset scale whose activation length is less than the first preset multiple and the updated aggregated power data of the preset scale whose ratio of the aggregated power point to the maximum power of the electrical appliance is greater than the third preset multiple and whose activation length is greater than the second preset multiple are removed to obtain the filtered aggregated power data set; wherein the first preset multiple is less than the second preset multiple, and the second preset multiple is less than the third preset multiple.

[0012] The present invention has the following beneficial effects: The present invention provides a sequence mapping load identification decomposition method based on a multi-scale residual neural network. The method constructs a load identification model with a network structure as a multi-scale residual network based on dilated convolution. The load identification model can avoid the common degradation problem that occurs when the traditional network increases the number of layers to learn more complex features. At the same time, the proposed dilated convolution can reduce a large number of model parameters and obtain a larger receptive field and a multi-scale data structure, so as to learn mixed data features more specifically, thereby enhancing the identification ability of the load identification model. Through the multi-scale structure, the model can achieve the best performance when facing electrical appliances with complex power characteristics. The aggregated power data set and the appliance-level power data set are preprocessed to obtain a preprocessed training set and a test set, which helps to improve the efficiency of the load identification model in the training process and further improve the identification performance of the load identification model. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0014] Figure 1 This is a flowchart of a method for identifying and decomposing sequence mapping loads based on a multi-scale residual neural network according to an embodiment of the present invention; Figure 2 is a schematic diagram of a load identification process based on NILM in an embodiment of the present invention; Figure 3 is a structural diagram of a dilated convolution residual block in an embodiment of the present invention; Figure 4 Schematic diagram of non-causal dilated convolution in an embodiment of the present invention; Figure 5 is an active power curve of a working cycle of a washing machine in an embodiment of the present invention; Figure 6 Schematic diagram of the overall structure of a multi-scale residual network based on dilated convolution in an embodiment of the present invention; Figure 7 This is the structural diagram of BiLSTM; Figure 8 This is the structural diagram of the denoising autoencoder; Fig. 9 This is the structural diagram of the traditional convolutional network; Fig.10 The results of testing different types of electrical appliances using different models in the embodiment of the present invention are shown in FIG. Figure 1 ; Fig.11 The results of testing different types of electrical appliances using different models in the embodiment of the present invention are shown in FIG. Figure 2 . DETAILED DESCRIPTION

[0015] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the technical solutions proposed by the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0016] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0017] Traditional load detection methods are based on event detection. The currently proposed latent factor Markov model is superior to the basic statistical model, but it may have some limitations when applied to complex implementation scenarios. When the amount of neural network knowledge data is large enough, it has strong nonlinear modeling capabilities, but due to the simple model structure, it lacks learning ability. Therefore, the identification effect of traditional load detection methods is poor.

[0018] In order to improve the recognition effect of load detection, this embodiment provides a sequence mapping load recognition decomposition method based on a multi-scale residual neural network. Figure 1 As shown, the following steps are included: S1, construct a load identification model with a multi-scale residual network based on dilated convolution.

[0019] In order to separate the active power data of each appliance from the total active power data, an independent neural network is trained for all types of appliances. The input of the neural network is the total power sequence, and the network outputs a sequence of corresponding length, which only contains the active power consumption of a single appliance. The schematic diagram of the load identification process based on NILM is shown in Figure 2 shown.

[0020] First, the dilated convolution residual block in the load identification model includes: This embodiment proposes to use a network architecture based on a dilated convolution residual block as a basic structural unit. The residual block can avoid the problem of gradient vanishing or exploding when the network layer increases. At the same time, the residual block can ensure that the network fully utilizes its depth to enhance and improve the overall learning performance. The structural diagram of the dilated convolution residual block is shown in FIG. Figure 3 shown.

[0021] exist Figure 3 In , the input of a single dilated convolutional residual block passes through two convolutional layers and their corresponding normalization layers, dropout layers, and nonlinear activation functions in sequence; then, the original input is added to the temporal output through a shortcut connection to form the final output of the dilated convolutional residual block. It is worth noting that if the number of input channels is different from the number of temporal output channels, a convolutional layer (with a kernel size of 1) is used to change the number of channels to be equal to the number of temporal output channels, and then they are added.

[0022] It should be noted that the shortcut connection allows each dilated convolution residual block to learn features directly from the original input, which greatly improves the learning ability of the network; connecting all the dilated convolution residual blocks in series to form a residual block aggregate can expand the depth of the residual network; compared with traditional convolution, the dilated convolution in the residual block has two additional features: first, due to the sequence-to-sequence characteristics, it is necessary to ensure that the sequence length before and after each convolution is the same; second, there will be holes (i.e., zero values) in the convolution kernel, and the number of these holes depends on the dilation rate of the layer. This feature enables the network to obtain a larger receptive field by selectively skipping certain parts of the input without changing the size of the convolution kernel or increasing parameters. For example, if d represents the dilation rate, d-1 values ​​will be skipped between every two actual volume extremes, that is, when d=1, the dilated convolution obviously degenerates into an ordinary 1D convolution.

[0023] In the dilated convolution residual block in this embodiment, the two convolutional layers share the same dilation rate, and as the network depth increases, the dilation rate of the residual block will increase exponentially, and the receptive field will increase at the same time. The schematic diagram of the non-causal dilated convolution is as follows: Figure 4 As shown, in Figure 4 The case with a kernel size of 3 is shown in Figure 2.

[0024] exist Figure 4 In is the original input sequence, is the final output sequence. It can be shown that for time series, convolution has a non-causal characteristic, that is, its output value at time t is determined by past and future information.

[0025] In this embodiment, when load identification is regarded as a denoising problem, more information from the sample context can help the network better learn the characteristics of pure signals. Unlike RNNs that are strictly calculated in sequence, i.e., RNNs will lead to a large consumption of training and prediction time, the dilated convolutional network has the advantage of processing multiple timelines at the same time, and the receptive field has a higher flexibility. In particular, if the dilation rate d of the last residual block is , then its receptive field can be calculated according to the following formula, which can be: ; In the formula, S represents the receptive field of the dilated convolution residual block, N represents the number of dilated convolutions, n represents the number of network layers of dilated convolutions, and k represents the convolution kernel size. When the convolution kernel size represented by k is an odd number, then the 2 in the calculation formula is n It represents the dilation rate in the residual block.

[0026] Secondly, regarding the multi-scale residual network structure in the load identification model, it includes: First, considering that different appliances have different working modes, even appliances of the same type may have different sub-modes within a working cycle. In different working modes and sub-modes, the power curve characteristics, including duration, power, and fluctuation, are significantly different from each other. For example, the active power curve of a washing machine working cycle is as follows: Figure 5 As shown, Figure 5 The working modes in the two dotted boxes are washing and auxiliary heating, and the power characteristics of these two modes are obviously different.

[0027] If a receptive field can only cover a specific working mode, then the network will have a better representation of that specific feature mode. However, if the receptive field is increased blindly, learning may become unclear, and the features of different working modes may be mixed and overlapped, thereby deteriorating the final recognition performance. Therefore, this embodiment proposes a new structure that can merge different receptive fields together so that the network can learn representations and combine the activation durations of different appliances from multiple receptive fields to reduce information loss. The new structure can be a multi-scale residual network structure. The schematic diagram of the overall structure of a multi-scale residual network based on dilated convolution is shown in FIG. Figure 6 shown.

[0028] exist Figure 6 In the figure, c represents the output channel, d represents the dilation rate, and k represents the convolution kernel size. The input sequence is fed into four residual blocks, each with a convolution kernel size of 5, which contain two blocks, three blocks, four blocks, and five blocks for feature extraction. In each residual block, the dilation rate and output channel of the first and last blocks are specified. All the blocks in the middle are multiplied in sequence. Based on the calculation formula of the receptive field, the receptive field contains 25, 57, 121, and 249 sampling points, respectively.

[0029] The dataset used in the verification process of this embodiment is the application-level electricity dataset (UK-DALE), and its sampling frequency is 6 seconds per point. Therefore, this embodiment can correspond the size of the receptive field to the actual time duration as 150 seconds, 342 seconds, 726 seconds and 1494 seconds; then, the output of each residual block is connected in depth, sent to two fully connected layers, and output through the sigmoid activation function; the activation function of all other layers is ReLU (Rectified Linear Unit), and the stride of all convolutional layers is 1.

[0030] It should be noted that the proposed new convolutional model of the residual module can avoid the common degradation problem of traditional networks when increasing the number of layers to learn more complex features; at the same time, the proposed dilated convolution can reduce the massive model parameters and obtain a larger receptive field and multi-scale data structure, so as to learn mixed data features more specifically.

[0031] So far, this embodiment has obtained a load identification model.

[0032] S2, obtaining an aggregated power dataset and an appliance-level power dataset from a public dataset, preprocessing the aggregated power dataset and the appliance-level power dataset to obtain a preprocessed training set and a test set.

[0033] The above step S2 can be implemented by the following steps S21 to S22: S21, obtaining an aggregated power dataset and an appliance-level power dataset from a public dataset.

[0034] The first step is to obtain the UK-DALE public data set, and select the aggregate power data and appliance-level power data of each target appliance in a first preset number of target households in the UK-DALE public data set according to the same sampling frequency.

[0035] Here, the first preset number of target households is a household with the second preset number of electrical appliances of the same type, and the target electrical appliances are electrical appliances of the same type. The first preset number can be a positive integer greater than or equal to 3, and the second preset number is 5.

[0036] In this embodiment, the dataset used is the public dataset UK-DALE, which records the appliance-level power data of five households. It contains a variety of appliances with different power levels to choose from. All appliance-level power data are active power sampled every 6 seconds, and the aggregated power data are apparent power sampled every 6 seconds, as well as active and reactive power sampled every 1 second. In order to facilitate subsequent model training, all active power is downsampled to a resolution of 6 seconds, and when no active power is available, only apparent power is selected.

[0037] Taking into account the sample size, power consumption ratio and representativeness of power characteristics, this embodiment selects the following five appliances for data analysis: kettle, refrigerator, washing machine, microwave oven and dishwasher. The five appliances are distributed in at least three households in UK-DALE. The specific distribution of appliances can be seen in Table 1.

[0038] Table 1 This embodiment uses the data of house No. 5 as the test set, and the data of other houses as the training set. In addition, since the washing machine and microwave oven in house No. 4 share the same electric meter, the power data of the washing machine and microwave oven in house No. 4 will not be used in the training.

[0039] The second step is to determine the time period when the appliances of each target household are turned on. Based on the obtained aggregate power dataset and appliance-level power dataset, the Electric.get_activation function of non-intrusive load monitoring and the parameters provided by UK-DALE are used to obtain activation information during the turn-on period, and then a pair of aggregate power sequences and label power sequences corresponding to each time index are obtained.

[0040] In this embodiment, in order to provide training samples and labels, it is necessary to first determine the time period when the appliance is turned on. The activation information is obtained using the Electric.get_activation function of NILMTK and the parameters provided by UK-DALE, which makes each time index have a pair of aggregated power sequences from the aggregated data. and tag power sequences from appliance-level data Since the length of each pair of sequences is different, a constant window length is set for each appliance. Then, training and test samples are generated by randomly placing windows under the premise of satisfying the entire activation interval. The parameters of the Electric.get_activation function are shown in Table 2: Table 2 The usage time of different electrical appliances is shown in Table 3: Table 3 S22, preprocessing the aggregated power data set and the appliance-level power data set to obtain a preprocessed training set and a test set.

[0041] In this embodiment, before starting to train the model, the data set is preprocessed as follows.

[0042] 1) First, the aggregate power dataset and the appliance-level power dataset are normalized to obtain the normalized aggregate power dataset and appliance-level power dataset.

[0043] In this embodiment, each pair of training samples and labels is normalized to facilitate network training. Specifically, all data in each pair are placed at the maximum power value of the pair of cluster power sequences, so that all values ​​range from 0 to 1. When generating the test set, a normalized data processing method is also used.

[0044] 2) Based on the normalized aggregated power dataset and appliance-level power dataset, a window of preset scale is used to divide the power data that does not contain the activation information of the target appliance from the activation starting point to obtain the updated aggregated power dataset and appliance-level power dataset. The updated appliance-level power dataset is used as the preprocessed test set.

[0045] In this embodiment, a window length is set in advance, and some training samples that do not contain activation of the target appliance are divided from the starting point of activation to expand the learning ability of the network. When generating the test set, the same data processing process is also used.

[0046] 3) Filter the updated aggregated power data set to obtain a filtered aggregated power data set as a preprocessed training set.

[0047] Specifically, the updated aggregated power data with a preset scale whose activation length is less than the first preset multiple and the updated aggregated power data with a preset scale whose ratio of the aggregated power point to the maximum power of the electrical appliance is greater than the third preset multiple and whose activation length is greater than the second preset multiple are removed to obtain the aggregated power data set after filtering. The first preset multiple is less than the second preset multiple, and the second preset multiple is less than the third preset multiple.

[0048] In this embodiment, the training samples are filtered, specifically: for samples whose activation length is less than 1 / 3 of the window length, and samples whose ratio of the aggregated power point to the maximum power of the electrical appliance is more than three times, and whose activation length exceeds half of the window length, they are discarded and not included in the training set. It should be noted that this rule will not be applied when generating the test set in order to obtain the most representative test results.

[0049] So far, this embodiment has obtained the preprocessed training set and test set for training the load identification model.

[0050] S3, using the preprocessed training set and test set to train and test the load identification model to obtain a trained load identification model.

[0051] In this embodiment, when training the load identification model, the optimizer used is the Adam optimizer, which can adaptively accelerate the convergence of the network; the loss function used in the training process is the cross entropy loss function, which shows better performance than the mean square error, which helps to improve the identification effect of the load identification model.

[0052] After obtaining the trained load identification model, the aggregated power data of the household to be identified can be directly input into the trained load identification model to obtain the load identification result of the household to be identified.

[0053] In order to verify the effectiveness of the sequence mapping load identification decomposition method based on multi-scale residual neural network in the present invention, a comparative analysis is conducted with three existing models, namely BiLSTM, denoising autoencoder, and traditional convolutional network. Figure 7 As shown, the structure diagram of the denoising self-encoder is as follows Figure 8 As shown in the figure, the structure of the traditional convolutional network is as follows Fig. 9 To improve the reliability of the verification results, the Adam optimizer is used for the traditional convolutional network; the stochastic gradient descent with Nesterov momentum is used for dBiLSTM and denoising autoencoder, with a momentum value of 0.9; all three existing models are trained using mean square error as the loss function.

[0054] In the verification process, the indicators for identification performance evaluation include: true positive (TP), false positive (FP), true negative (TN), and false negative (FN). The actual power of the appliance at time t is recorded as , the estimated power is recorded as , the calculation formulas of the four static evaluation indicators can be: ; ; ; ; In the formula, Recall represents the recall rate, Precision represents the precision rate, F1 Score represents the F1 score, MeanAbsolute Error represents the mean absolute error, T represents the time period, Represents the absolute value function.

[0055] Among them, recall and precision are usually regarded as intermediate indicators, because even high scores on these two indicators may lead to a poor F1 score, which can reflect the overall performance of the network in classifying the switching status of electrical appliances; on the other hand, the mean absolute error (MAE) reflects the performance of the network in more specific power deviations.

[0056] The verification process was carried out in the experimental environment shown in Table 4. The results of testing different types of appliances using different models are shown in Fig.10 and Fig.11 in Fig.10 In the figure, a represents the test result of the electric kettle, b represents the test result of the microwave oven, c represents the test result of the refrigerator, d represents the test result of the dishwasher, and e represents the test result of the washing machine. The number of parameters of each model is counted using Tensorflow Profiler, and the time.time() function is used to measure the prediction time of each sample. The number of parameters and time results of the model are shown in Tables 5 and 6, respectively.

[0057] Table 4 Table 5 Table 6 All MAE results of each appliance are divided by the maximum MAE value of the corresponding appliance so that they are normalized to the [0, 1] interval and clearly displayed in the figure. The denominator values ​​are shown in Table 7: Table 7 Appliances with simple power features, such as kettles and dishwashers, generally show better recognition performance than other appliances, while appliances with more difficult to capture features, such as microwave ovens, generally show poor recognition performance. Specifically for each indicator, the kettle has the best F1 score among the five models, but the MAE is poor. Because the activation of the kettle is concentrated in a smaller window and the power is higher (over 2000W), small errors will be amplified to a high level after the output of the neural network is activated by sigmoid. Similarly, the MAE performance of the refrigerator (lower power) is better. The washing machine and dishwasher have a lower power threshold, so there are fewer false negative or false positive points, resulting in the recall rate being generally higher than the precision, while more comprehensive indicators, such as F1 and MAE, are different.

[0058] From the verification results, it can be seen that the traditional convolutional network performs better than the autoencoder on the kettle and washing machine, but not on the dishwasher. The performance of BiLSTM is mediocre, with a large MAE and occasionally a good F1 score on three appliances. At the same time, the proposed multi-scale model performs better on all five appliances. Comparing the residual network based on dilated convolution with other networks based on CNN (Convolutional Neural Network, convolutional neural network), its F1 and MAE results on five appliances surpassed those of other models. It can be seen that the residual network and dilated convolution can solve the problems of insufficient receptive field and performance degradation in ordinary convolution. Therefore, the sequence mapping load identification decomposition method based on multi-scale residual neural network proposed in the present invention is faster and better than RNN in identification performance, and compared with other CNN-based models, it reduces the number of parameters while also reducing time costs.

[0059] In summary, the network structure proposed in this embodiment is a load identification model based on a multi-scale residual network with dilated convolution. Compared with the traditional model, the dilated convolution and multi-scale structure of this load identification model can be successfully applied to time series load identification, and can effectively solve the NILM problem; and the large receptive field generated by the dilated convolution can obtain more information from the background to assist in identification, while the multi-scale structure enables the model to achieve optimal performance when facing electrical appliances with complex power characteristics; in addition, this embodiment shows good potential in terms of small size, which helps to shorten the time used for prediction. In subsequent progress, this load identification model can be extended to the sequence-to-point regression task to directly estimate the energy consumption after identification.

[0060] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit them. A person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A sequence mapping load identification and decomposition method based on multi-scale residual neural network, characterized in that: The following steps are involved: Constructing a load identification model, wherein the network structure of the load identification model is a multi-scale residual network based on dilated convolution; Obtaining an aggregated power data set and an appliance-level power data set from a public data set, preprocessing the aggregated power data set and the appliance-level power data set to obtain a preprocessed training set and a test set; The load identification model is trained and tested using the preprocessed training set and test set to obtain a trained load identification model.

2. The method for sequence mapping load identification and decomposition based on multi-scale residual neural network according to claim 1 is characterized in that: The multi-scale residual network based on dilated convolution includes: a dilated convolution residual block; The input of the dilated convolution residual block passes through two convolution layers and their corresponding normalization layers, dropout layers, and nonlinear activation functions in sequence; then, the original input is added to the temporal output through a shortcut connection to form the final output of the dilated convolution residual block; wherein the two convolution layers in the dilated convolution residual block share the same dilation rate.

3. The method for sequence mapping load identification and decomposition based on multi-scale residual neural network according to claim 2 is characterized in that: The calculation formula of the receptive field of the dilated convolution residual block is: ； Where S represents the receptive field of the dilated convolution residual block, N represents the number of dilated convolutions, n represents the number of network layers of dilated convolutions, and k represents the convolution kernel size.

4. The method for sequence mapping load identification and decomposition based on multi-scale residual neural network according to claim 1, characterized in that: The obtaining of the aggregated power dataset and the appliance-level power dataset in the public dataset includes: Obtain a UK-DALE public data set, and select aggregate power data and appliance-level power data of each target appliance in a first preset number of target households according to the same sampling frequency in the UK-DALE public data set; wherein the first preset number of target households are households with a second preset number of appliances of the same type, and the target appliances are appliances of the same type; Determine the time period when the electrical appliances of each target household are turned on. Based on the obtained aggregate power dataset and appliance-level power dataset, use the Electric.get_activation function of non-intrusive load monitoring and the parameters provided by UK-DALE to obtain activation information during the turn-on period, and then obtain a pair of aggregate power sequences and label power sequences corresponding to each time index.

5. The method for sequence mapping load identification and decomposition based on multi-scale residual neural network according to claim 1 is characterized in that: The preprocessing of the aggregated power data set and the appliance-level power data set to obtain a preprocessed training set and a test set includes: First, normalizing the aggregated power data set and the appliance-level power data set to obtain a normalized aggregated power data set and an appliance-level power data set; Based on the normalized aggregate power data set and appliance-level power data set, a window of preset scale is used to divide the power data that does not contain the activation information of the target appliance from the activation starting point, and the updated aggregate power data set and appliance-level power data set are obtained. The updated appliance-level power data set is used as the preprocessed test set. The updated aggregated power data set is filtered to obtain a filtered aggregated power data set as a preprocessed training set.

6. The method for sequence mapping load identification and decomposition based on multi-scale residual neural network according to claim 5 is characterized in that: The filtering the updated aggregate power data set to obtain the filtered aggregate power data set includes: The updated aggregated power data of the preset scale whose activation length is less than the first preset multiple and the updated aggregated power data of the preset scale whose ratio of the aggregated power point to the maximum power of the electrical appliance is greater than the third preset multiple and whose activation length is greater than the second preset multiple are removed to obtain the filtered aggregated power data set; wherein the first preset multiple is less than the second preset multiple, and the second preset multiple is less than the third preset multiple.

Images